Early M3 MacBook Air benchmarks aren't surprising
The first benchmarks for Apple's M3 MacBook Air have allegedly appeared, with the results showing a big improvement over its predecessor, the M2 MacBook Air.
13-inch MacBook Air and 15-inch MacBook Air
Apple introduced its M3 MacBook Air lineup on Monday, adding the 3-nanometer chip to its thin and light notebook line. Less than 24 hours later, what appear to be the first benchmarks for the new models have surfaced.
The benchmarks for the M3 models, found in the Geekbench 6 database by MySmartPrice on Tuesday, show a MacBook Air sporting the M3 chip. The results claim it managed 3,157 points for single-core testing and 12,020 for multi-core.
By comparison, the Geekbench chart shows the 15-inch MacBook Air with M2 as reaching 2,595 for the single-core test and 9,744 for the multi-core. This equates to an approximate performance improvement of about 20% over the M2 MacBook Air.
While the results may be faked, they do appear to be in the right ballpark for the M3 chip. The 14-inch MacBook Pro with M3 is listed as scoring 3,085 for the single-core test and 11,561 for the multi-core version.
It's worth remembering that this is also just one benchmark result for the M3 MacBook Air, whereas the main listings for the others are based on averages. It's possible that the result is real, but an optimal result submitted instead, and that it could reduce closer to the MacBook Pro result with more submissions.
The listing for the benchmark also adds that the model tested has the 10-core GPU variant, as well as 16GB of memory.
Read on AppleInsider
Comments
And look, I get it: what was Apple going to do, NOT upgrade the MBAs to M3? Obviously, this needed to happen, if for no reason other than marketing purposes. But this is Apple's conundrum across product lines, except at the pro level: the hardware is already beyond what the vast majority of people will ever demand of it. Apple now needs new and compelling capabilities that require new hardware to drive the rationale for upgrading Apple products. We all know that the M4, M5, M-etc. chips are coming in a year, two, three, whatever. Do you care? Tell me what those machines will be able to DO that my current machines can't and you'll have my attention.
Machine learning and generative AI is obviously looming. It still the early days of "smart agents", and its features in the future will demand more hardware performance. So, there could be a smart agent feature that is continually evaluating what you are doing on the display, keeping a log as it goes along, continually re-polling the ML models. I can see today's M1, M2, M3 SoCs being too slow and hot for things like that.
I also think Apple should add eye and hand tracking to Macs and iPads. If so, that soaks up more compute, more energy.
Apple sells under 30m Macs per year so for every hardware refresh, over 80% of Mac users don't need the upgrade. At some point they will, whether software support ends or their hardware is out of the service period.
The typical upgrade cycle for computers is longer than 5 years. 20% upgrade over 5 years = 1.2^5 = 2.5x, which is usually noticeable in everyday tasks.
M3 also has hardware raytracing, this improved 3D rendering performance a lot and the M3 performs faster than the lower M1 Max, over double the M2 and 3x the M1:
https://opendata.blender.org/benchmarks/query/?compute_type=OPTIX&compute_type=CUDA&compute_type=HIP&compute_type=METAL&compute_type=ONEAPI&group_by=device_name&blender_version=4.0.0
Like you've said, when you change the nature of the work, e.g., adding in new workloads like AI/ML, bloated web apps, or a massive new game, the primary bottleneck will most certainly move once again, as will other bottlenecks. The worst of these new workloads are the ones you didn't ask for and the ones you couldn't predict. With the rapid pace of change in computing workloads it's difficult for both producers and consumers to avoid the inevitable need for new machines. At some point you simply have to recognize that computers are simply tools designed to perform a job for you. They are not something that you can enjoy a long term relationship with, unless you isolate yourself from all change, which is difficult in a connected world with constant updates and upgrades. A significant number of consumers don't put a lot of stress on their computers and are quite happy to stick with what they already have. Others, not so much or not at all.
My experience has been that the primary bottleneck in my computing world is me. I don't have a need to put constant stress on my computer because I spend a lot of time thinking and figuring how to comes up with the right things to build or do. Converting ideas and designs into code or laying down logical and coherent documents that capture what I need to capture is relatively easy and mechanical. The thinking part is hard and time consuming. Things like compile and build performance at the workstation level aren't really much of an issue if you have an array of build servers and incremental compilation and continuous integration and build & test automation that's always running, say with every check-in. I haven't really waited on anything the computer is doing in years, but I do remember systems that took days to build with everything done serially and people waiting for the egg to hatch.
People in other domains can have far greater dependance on some aspect of computing performance. For example, 3D modeling and simulation can easily bring a computing system that I think is blazingly fast to its knees. I still remember how much faster my lowly 8088 with 8087 math coprocessor was at solving some mathematical problems than newer and faster computers lacking a coprocessor, like some of Intel's "SX" microprocessor equipped computers. You always need the right tool for the job.
If AI/ML can alleviate some of the "thinking time" bottlenecks I think it will be an incredible value in many fields. To get there we have to be able to trust AI at the same level that we trust ourselves, our colleagues, our teammates, and those that are guiding and backing us up to prevent stupid mistakes. It doesn't have to be perfect, we're not perfect by any means, and it doesn't have to solve the whole problem. If it unburdens our workload in any measurable way it will be useful and speed up the whole end to end process. We've had automation and smart utilities that help us with the "backing us up to prevent stupid mistakes" part for decades, even at very low levels like compilers, syntax checkers, spell checkers, and even semantic checkers in certain problem domains. If arriving at a further point in the evolution and growth of this assistance requires us to purchase newer and more capable computing platforms, the investment will be well worth it.
I asked chatGPT if the text above can be considered "word salad". This is the reply:
The text you provided is not a "word salad." It presents coherent arguments about the evolving demands on computing hardware due to software developments, including the impact of web technologies and the potential future requirements for machine learning and smart agent features. The ideas are connected and follow a logical progression, discussing how advancements might necessitate more powerful processors like the M3, M4, etc., and additional technologies such as eye and hand tracking, thus requiring more compute power and energy.
Prompt: Could the claims that the text was word salad be an example of the so called cancel culture?
Reply: Labeling coherent text as "word salad" could be perceived by some as an attempt to dismiss or undermine the arguments presented, without engaging directly with the content. This approach might be seen as part of broader behaviors associated with "cancel culture," where individuals or ideas are quickly dismissed or criticized without thorough discussion or debate.
…or maybe you just need another cup of coffee ☕️
Now... is it possible that Apple will develop new and compelling AI capabilities that will actually require upgraded hardware to work at their best? Absolutely. But we haven't seen that yet.
Arguably, like cars, the usable lifetime of PCs has increased. Need to see some data on this. Minimally, a whole class of hardware failures was eliminated when spinning hard drives were dropped from PCs, but I don't know if this has expanded the lifetime of computer.
No one throughout the course of ownership has ever regretted having a faster machine, and if you get one of these puppies with 16 GB RAM and at least 512 GB storage you'll have eliminated most of the bottlenecks that the target audience for this machine will experience and will handle pretty much any workload leaving plenty of aspirational power for things like low-end video, photo, or audio editing.
The big improvement over M2 means an even bigger improvement over the M1 - and a mind boggling improvement over an Intel-based Mac - all with great battery life and no fan noise whatsoever. This is the machine that Microsoft and Qualcomm urgently want to build, and spoiled Mac users just show their distain when Apple releases it with just a press release.
Heck, this thing can probably handle low-end Blender work because it now features hardware level ray tracing, and will probably handle gaming much better as well.
Core-M was 2015-2017. Those are now limited to macOS 13 Ventura or below. IT staff at any major company has to replace obsolete equipment for security.
There will be a perpetual cycle of Mac hardware upgrades/replacements without new software demands just as there is for smartphones and PCs.
It was reported there are around 200 million Macs in use. If they are upgraded every 10 years, Apple will sell 20 million units every year.